Create.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
   5  *
   6  *
   7  *    This program is free software; you can redistribute it and/or modify
   8  *    it under the terms of the GNU General Public License as published by
   9  *    the Free Software Foundation; either version 2 of the License, or
  10  *    (at your option) any later version.
  11  *
  12  *    This program is distributed in the hope that it will be useful,
  13  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *    GNU General Public License for more details.
  16  *
  17  *    You should have received a copy of the GNU General Public License
  18  *    along with this program; if not, write to the Free Software
  19  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  *
  21  *    Author: Neil Brown
  22  *    Email: <neilb@suse.de>
  23  */
  24
  25 #include "mdadm.h"
  26 #include        "md_u.h"
  27 #include        "md_p.h"
  28 #include        <ctype.h>
  29
  30 static int default_layout(struct supertype *st, int level, int verbose)
  31 {
  32         int layout = UnSet;
  33
  34         if (st && st->ss->default_geometry)
  35                 st->ss->default_geometry(st, &level, &layout, NULL);
  36
  37         if (layout == UnSet)
  38                 switch(level) {
  39                 default: /* no layout */
  40                         layout = 0;
  41                         break;
  42                 case 10:
  43                         layout = 0x102; /* near=2, far=1 */
  44                         if (verbose > 0)
  45                                 fprintf(stderr,
  46                                         Name ": layout defaults to n2\n");
  47                         break;
  48                 case 5:
  49                 case 6:
  50                         layout = map_name(r5layout, "default");
  51                         if (verbose > 0)
  52                                 fprintf(stderr,
  53                                         Name ": layout defaults to %s\n", map_num(r5layout, layout));
  54                         break;
  55                 case LEVEL_FAULTY:
  56                         layout = map_name(faultylayout, "default");
  57
  58                         if (verbose > 0)
  59                                 fprintf(stderr,
  60                                         Name ": layout defaults to %s\n", map_num(faultylayout, layout));
  61                         break;
  62                 }
  63
  64         return layout;
  65 }
  66
  67
  68 int Create(struct supertype *st, char *mddev,
  69            int chunk, int level, int layout, unsigned long long size,
  70            int raiddisks, int sparedisks,
  71            char *name, char *homehost, int *uuid,
  72            int subdevs, struct mddev_dev *devlist,
  73            int runstop, int verbose, int force, int assume_clean,
  74            char *bitmap_file, int bitmap_chunk, int write_behind,
  75            int delay, int autof)
  76 {
  77         /*
  78          * Create a new raid array.
  79          *
  80          * First check that necessary details are available
  81          * (i.e. level, raid-disks)
  82          *
  83          * Then check each disk to see what might be on it
  84          * and report anything interesting.
  85          *
  86          * If anything looks odd, and runstop not set,
  87          * abort.
  88          *
  89          * SET_ARRAY_INFO and ADD_NEW_DISK, and
  90          * if runstop==run, or raiddisks disks were used,
  91          * RUN_ARRAY
  92          */
  93         int mdfd;
  94         unsigned long long minsize=0, maxsize=0;
  95         char *mindisc = NULL;
  96         char *maxdisc = NULL;
  97         int dnum;
  98         struct mddev_dev *dv;
  99         int fail=0, warn=0;
 100         struct stat stb;
 101         int first_missing = subdevs * 2;
 102         int second_missing = subdevs * 2;
 103         int missing_disks = 0;
 104         int insert_point = subdevs * 2; /* where to insert a missing drive */
 105         int total_slots;
 106         int pass;
 107         int vers;
 108         int rv;
 109         int bitmap_fd;
 110         int have_container = 0;
 111         int container_fd = -1;
 112         int need_mdmon = 0;
 113         unsigned long long bitmapsize;
 114         struct mdinfo info, *infos;
 115         int did_default = 0;
 116         int do_default_layout = 0;
 117         int do_default_chunk = 0;
 118         unsigned long safe_mode_delay = 0;
 119         char chosen_name[1024];
 120         struct map_ent *map = NULL;
 121         unsigned long long newsize;
 122
 123         int major_num = BITMAP_MAJOR_HI;
 124
 125         memset(&info, 0, sizeof(info));
 126         if (level == UnSet && st && st->ss->default_geometry)
 127                 st->ss->default_geometry(st, &level, NULL, NULL);
 128         if (level == UnSet) {
 129                 fprintf(stderr,
 130                         Name ": a RAID level is needed to create an array.\n");
 131                 return 1;
 132         }
 133         if (raiddisks < 4 && level == 6) {
 134                 fprintf(stderr,
 135                         Name ": at least 4 raid-devices needed for level 6\n");
 136                 return 1;
 137         }
 138         if (raiddisks > 256 && level == 6) {
 139                 fprintf(stderr,
 140                         Name ": no more than 256 raid-devices supported for level 6\n");
 141                 return 1;
 142         }
 143         if (raiddisks < 2 && level >= 4) {
 144                 fprintf(stderr,
 145                         Name ": at least 2 raid-devices needed for level 4 or 5\n");
 146                 return 1;
 147         }
 148         if (level <= 0 && sparedisks) {
 149                 fprintf(stderr,
 150                         Name ": This level does not support spare devices\n");
 151                 return 1;
 152         }
 153
 154         if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) {
 155                 /* If given a single device, it might be a container, and we can
 156                  * extract a device list from there
 157                  */
 158                 mdu_array_info_t inf;
 159                 int fd;
 160
 161                 memset(&inf, 0, sizeof(inf));
 162                 fd = open(devlist->devname, O_RDONLY);
 163                 if (fd >= 0 &&
 164                     ioctl(fd, GET_ARRAY_INFO, &inf) == 0 &&
 165                     inf.raid_disks == 0) {
 166                         /* yep, looks like a container */
 167                         if (st) {
 168                                 rv = st->ss->load_container(st, fd,
 169                                                             devlist->devname);
 170                                 if (rv == 0)
 171                                         have_container = 1;
 172                         } else {
 173                                 st = super_by_fd(fd, NULL);
 174                                 if (st && !(rv = st->ss->
 175                                             load_container(st, fd,
 176                                                            devlist->devname)))
 177                                         have_container = 1;
 178                                 else
 179                                         st = NULL;
 180                         }
 181                         if (have_container) {
 182                                 subdevs = raiddisks;
 183                                 first_missing = subdevs * 2;
 184                                 second_missing = subdevs * 2;
 185                                 insert_point = subdevs * 2;
 186                         }
 187                 }
 188                 if (fd >= 0)
 189                         close(fd);
 190         }
 191         if (st && st->ss->external && sparedisks) {
 192                 fprintf(stderr,
 193                         Name ": This metadata type does not support "
 194                         "spare disks at create time\n");
 195                 return 1;
 196         }
 197         if (subdevs > raiddisks+sparedisks) {
 198                 fprintf(stderr, Name ": You have listed more devices (%d) than are in the array(%d)!\n", subdevs, raiddisks+sparedisks);
 199                 return 1;
 200         }
 201         if (!have_container && subdevs < raiddisks+sparedisks) {
 202                 fprintf(stderr, Name ": You haven't given enough devices (real or missing) to create this array\n");
 203                 return 1;
 204         }
 205         if (bitmap_file && level <= 0) {
 206                 fprintf(stderr, Name ": bitmaps not meaningful with level %s\n",
 207                         map_num(pers, level)?:"given");
 208                 return 1;
 209         }
 210
 211         /* now set some defaults */
 212
 213
 214         if (layout == UnSet) {
 215                 do_default_layout = 1;
 216                 layout = default_layout(st, level, verbose);
 217         }
 218
 219         if (level == 10)
 220                 /* check layout fits in array*/
 221                 if ((layout&255) * ((layout>>8)&255) > raiddisks) {
 222                         fprintf(stderr, Name ": that layout requires at least %d devices\n",
 223                                 (layout&255) * ((layout>>8)&255));
 224                         return 1;
 225                 }
 226
 227         switch(level) {
 228         case 4:
 229         case 5:
 230         case 10:
 231         case 6:
 232         case 0:
 233                 if (chunk == 0 || chunk == UnSet) {
 234                         chunk = UnSet;
 235                         do_default_chunk = 1;
 236                         /* chunk will be set later */
 237                 }
 238                 break;
 239         case LEVEL_LINEAR:
 240                 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
 241                 if (get_linux_version() < 2006016 && chunk == 0) {
 242                         chunk = 64;
 243                         if (verbose > 0)
 244                                 fprintf(stderr, Name ": chunk size defaults to 64K\n");
 245                 }
 246                 break;
 247         case 1:
 248         case LEVEL_FAULTY:
 249         case LEVEL_MULTIPATH:
 250         case LEVEL_CONTAINER:
 251                 if (chunk) {
 252                         chunk = 0;
 253                         if (verbose > 0)
 254                                 fprintf(stderr, Name ": chunk size ignored for this level\n");
 255                 }
 256                 break;
 257         default:
 258                 fprintf(stderr, Name ": unknown level %d\n", level);
 259                 return 1;
 260         }
 261
 262         if (size && chunk && chunk != UnSet)
 263                 size &= ~(unsigned long long)(chunk - 1);
 264         newsize = size * 2;
 265         if (st && ! st->ss->validate_geometry(st, level, layout, raiddisks,
 266                                               &chunk, size*2, NULL, &newsize, verbose>=0))
 267                 return 1;
 268
 269         if (chunk) {
 270                 newsize &= ~(unsigned long long)(chunk*2 - 1);
 271                 size &= ~(unsigned long long)(chunk - 1);
 272         }
 273         if (size == 0) {
 274                 size = newsize / 2;
 275                 if (size && verbose > 0)
 276                         fprintf(stderr, Name ": setting size to %lluK\n",
 277                                 (unsigned long long)size);
 278         }
 279
 280         /* now look at the subdevs */
 281         info.array.active_disks = 0;
 282         info.array.working_disks = 0;
 283         dnum = 0;
 284         for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) {
 285                 char *dname = dv->devname;
 286                 unsigned long long freesize;
 287                 if (strcasecmp(dname, "missing")==0) {
 288                         if (first_missing > dnum)
 289                                 first_missing = dnum;
 290                         if (second_missing > dnum && dnum > first_missing)
 291                                 second_missing = dnum;
 292                         missing_disks ++;
 293                         continue;
 294                 }
 295                 info.array.working_disks++;
 296                 if (dnum < raiddisks)
 297                         info.array.active_disks++;
 298                 if (st == NULL) {
 299                         struct createinfo *ci = conf_get_create_info();
 300                         if (ci)
 301                                 st = ci->supertype;
 302                 }
 303                 if (st == NULL) {
 304                         /* Need to choose a default metadata, which is different
 305                          * depending on geometry of array.
 306                          */
 307                         int i;
 308                         char *name = "default";
 309                         for(i=0; !st && superlist[i]; i++) {
 310                                 st = superlist[i]->match_metadata_desc(name);
 311                                 if (do_default_layout)
 312                                         layout = default_layout(st, level, verbose);
 313                                 if (st && !st->ss->validate_geometry
 314                                                 (st, level, layout, raiddisks,
 315                                                  &chunk, size*2, dname, &freesize,
 316                                                  verbose > 0)) {
 317                                         free(st);
 318                                         st = NULL;
 319                                         chunk = do_default_chunk ? 0 : chunk;
 320                                 }
 321                         }
 322
 323                         if (!st) {
 324                                 fprintf(stderr, Name ": device %s not suitable "
 325                                         "for any style of array\n",
 326                                         dname);
 327                                 exit(2);
 328                         }
 329                         if (st->ss != &super0 ||
 330                             st->minor_version != 90)
 331                                 did_default = 1;
 332                 } else {
 333                         if (do_default_layout)
 334                                 layout = default_layout(st, level, verbose);
 335                         if (!st->ss->validate_geometry(st, level, layout,
 336                                                        raiddisks,
 337                                                        &chunk, size*2, dname,
 338                                                        &freesize,
 339                                                        verbose >= 0)) {
 340
 341                                 fprintf(stderr,
 342                                         Name ": %s is not suitable for "
 343                                         "this array.\n",
 344                                         dname);
 345                                 fail = 1;
 346                                 continue;
 347                         }
 348                 }
 349                 if (verbose > 0 && do_default_chunk) {
 350                         do_default_chunk = 0;
 351                         fprintf(stderr, Name ": chunk size "
 352                                 "defaults to %dK\n", chunk);
 353                 }
 354
 355                 freesize /= 2; /* convert to K */
 356                 if (chunk) {
 357                         /* round to chunk size */
 358                         freesize = freesize & ~(chunk-1);
 359                 }
 360
 361                 if (size && freesize < size) {
 362                         fprintf(stderr, Name ": %s is smaller than given size."
 363                                 " %lluK < %lluK + metadata\n",
 364                                 dname, freesize, size);
 365                         fail = 1;
 366                         continue;
 367                 }
 368                 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
 369                         maxdisc = dname;
 370                         maxsize = freesize;
 371                 }
 372                 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
 373                         mindisc = dname;
 374                         minsize = freesize;
 375                 }
 376                 if (runstop != 1 || verbose >= 0) {
 377                         int fd = open(dname, O_RDONLY);
 378                         if (fd <0 ) {
 379                                 fprintf(stderr, Name ": Cannot open %s: %s\n",
 380                                         dname, strerror(errno));
 381                                 fail=1;
 382                                 continue;
 383                         }
 384                         warn |= check_ext2(fd, dname);
 385                         warn |= check_reiser(fd, dname);
 386                         warn |= check_raid(fd, dname);
 387                         if (strcmp(st->ss->name, "1.x") == 0 &&
 388                             st->minor_version >= 1)
 389                                 /* metadata at front */
 390                                 warn |= check_partitions(fd, dname, 0, 0);
 391                         else if (level == 1 || level == LEVEL_CONTAINER
 392                                     || (level == 0 && raiddisks == 1))
 393                                 /* partitions could be meaningful */
 394                                 warn |= check_partitions(fd, dname, freesize*2, size*2);
 395                         else
 396                                 /* partitions cannot be meaningful */
 397                                 warn |= check_partitions(fd, dname, 0, 0);
 398                         if (strcmp(st->ss->name, "1.x") == 0 &&
 399                             st->minor_version >= 1 &&
 400                             did_default &&
 401                             level == 1 &&
 402                             (warn & 1024) == 0) {
 403                                 warn |= 1024;
 404                                 fprintf(stderr, Name ": Note: this array has metadata at the start and\n"
 405                                         "    may not be suitable as a boot device.  If you plan to\n"
 406                                         "    store '/boot' on this device please ensure that\n"
 407                                         "    your boot-loader understands md/v1.x metadata, or use\n"
 408                                         "    --metadata=0.90\n");
 409                         }
 410                         close(fd);
 411                 }
 412         }
 413         if (raiddisks + sparedisks > st->max_devs) {
 414                 fprintf(stderr, Name ": Too many devices:"
 415                         " %s metadata only supports %d\n",
 416                         st->ss->name, st->max_devs);
 417                 return 1;
 418         }
 419         if (have_container)
 420                 info.array.working_disks = raiddisks;
 421         if (fail) {
 422                 fprintf(stderr, Name ": create aborted\n");
 423                 return 1;
 424         }
 425         if (size == 0) {
 426                 if (mindisc == NULL && !have_container) {
 427                         fprintf(stderr, Name ": no size and no drives given - aborting create.\n");
 428                         return 1;
 429                 }
 430                 if (level > 0 || level == LEVEL_MULTIPATH
 431                     || level == LEVEL_FAULTY
 432                     || st->ss->external ) {
 433                         /* size is meaningful */
 434                         if (!st->ss->validate_geometry(st, level, layout,
 435                                                        raiddisks,
 436                                                        &chunk, minsize*2,
 437                                                        NULL, NULL, 0)) {
 438                                 fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
 439                                 return 1;
 440                         }
 441                         size = minsize;
 442                         if (verbose > 0)
 443                                 fprintf(stderr, Name ": size set to %lluK\n", size);
 444                 }
 445         }
 446         if (!have_container && level > 0 && ((maxsize-size)*100 > maxsize)) {
 447                 if (runstop != 1 || verbose >= 0)
 448                         fprintf(stderr, Name ": largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
 449                                 maxdisc, size);
 450                 warn = 1;
 451         }
 452
 453         if (st->ss->detail_platform && st->ss->detail_platform(0, 1) != 0) {
 454                 if (runstop != 1 || verbose >= 0)
 455                         fprintf(stderr, Name ": %s unable to enumerate platform support\n"
 456                                 "    array may not be compatible with hardware/firmware\n",
 457                                 st->ss->name);
 458                 warn = 1;
 459         }
 460
 461         if (warn) {
 462                 if (runstop!= 1) {
 463                         if (!ask("Continue creating array? ")) {
 464                                 fprintf(stderr, Name ": create aborted.\n");
 465                                 return 1;
 466                         }
 467                 } else {
 468                         if (verbose > 0)
 469                                 fprintf(stderr, Name ": creation continuing despite oddities due to --run\n");
 470                 }
 471         }
 472
 473         /* If this is raid4/5, we want to configure the last active slot
 474          * as missing, so that a reconstruct happens (faster than re-parity)
 475          * FIX: Can we do this for raid6 as well?
 476          */
 477         if (st->ss->external == 0 &&
 478             assume_clean==0 && force == 0 && first_missing >= raiddisks) {
 479                 switch ( level ) {
 480                 case 4:
 481                 case 5:
 482                         insert_point = raiddisks-1;
 483                         sparedisks++;
 484                         info.array.active_disks--;
 485                         missing_disks++;
 486                         break;
 487                 default:
 488                         break;
 489                 }
 490         }
 491         /* For raid6, if creating with 1 missing drive, make a good drive
 492          * into a spare, else the create will fail
 493          */
 494         if (assume_clean == 0 && force == 0 && first_missing < raiddisks &&
 495             st->ss->external == 0 &&
 496             second_missing >= raiddisks && level == 6) {
 497                 insert_point = raiddisks - 1;
 498                 if (insert_point == first_missing)
 499                         insert_point--;
 500                 sparedisks ++;
 501                 info.array.active_disks--;
 502                 missing_disks++;
 503         }
 504
 505         if (level <= 0 && first_missing < subdevs * 2) {
 506                 fprintf(stderr,
 507                         Name ": This level does not support missing devices\n");
 508                 return 1;
 509         }
 510
 511         /* We need to create the device */
 512         map_lock(&map);
 513         mdfd = create_mddev(mddev, name, autof, LOCAL, chosen_name);
 514         if (mdfd < 0)
 515                 return 1;
 516         mddev = chosen_name;
 517
 518         vers = md_get_version(mdfd);
 519         if (vers < 9000) {
 520                 fprintf(stderr, Name ": Create requires md driver version 0.90.0 or later\n");
 521                 goto abort;
 522         } else {
 523                 mdu_array_info_t inf;
 524                 memset(&inf, 0, sizeof(inf));
 525                 ioctl(mdfd, GET_ARRAY_INFO, &inf);
 526                 if (inf.working_disks != 0) {
 527                         fprintf(stderr, Name ": another array by this name"
 528                                 " is already running.\n");
 529                         goto abort;
 530                 }
 531         }
 532
 533         /* Ok, lets try some ioctls */
 534
 535         info.array.level = level;
 536         info.array.size = size;
 537         info.array.raid_disks = raiddisks;
 538         /* The kernel should *know* what md_minor we are dealing
 539          * with, but it chooses to trust me instead. Sigh
 540          */
 541         info.array.md_minor = 0;
 542         if (fstat(mdfd, &stb)==0)
 543                 info.array.md_minor = minor(stb.st_rdev);
 544         info.array.not_persistent = 0;
 545
 546         if ( ( (level == 4 || level == 5) &&
 547                (insert_point < raiddisks || first_missing < raiddisks) )
 548              ||
 549              ( level == 6 && (insert_point < raiddisks
 550                               || second_missing < raiddisks))
 551              ||
 552              ( level <= 0 )
 553              ||
 554              assume_clean
 555                 ) {
 556                 info.array.state = 1; /* clean, but one+ drive will be missing*/
 557                 info.resync_start = MaxSector;
 558         } else {
 559                 info.array.state = 0; /* not clean, but no errors */
 560                 info.resync_start = 0;
 561         }
 562         if (level == 10) {
 563                 /* for raid10, the bitmap size is the capacity of the array,
 564                  * which is array.size * raid_disks / ncopies;
 565                  * .. but convert to sectors.
 566                  */
 567                 int ncopies = ((layout>>8) & 255) * (layout & 255);
 568                 bitmapsize = (unsigned long long)size * raiddisks / ncopies * 2;
 569 /*              printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
 570         } else
 571                 bitmapsize = (unsigned long long)size * 2;
 572
 573         /* There is lots of redundancy in these disk counts,
 574          * raid_disks is the most meaningful value
 575          *          it describes the geometry of the array
 576          *          it is constant
 577          * nr_disks is total number of used slots.
 578          *          it should be raid_disks+spare_disks
 579          * spare_disks is the number of extra disks present
 580          *          see above
 581          * active_disks is the number of working disks in
 582          *          active slots. (With raid_disks)
 583          * working_disks is the total number of working disks,
 584          *          including spares
 585          * failed_disks is the number of disks marked failed
 586          *
 587          * Ideally, the kernel would keep these (except raid_disks)
 588          * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
 589          * So for now, we assume that all raid and spare
 590          * devices will be given.
 591          */
 592         info.array.spare_disks=sparedisks;
 593         info.array.failed_disks=missing_disks;
 594         info.array.nr_disks = info.array.working_disks
 595                 + info.array.failed_disks;
 596         info.array.layout = layout;
 597         info.array.chunk_size = chunk*1024;
 598
 599         if (name == NULL || *name == 0) {
 600                 /* base name on mddev */
 601                 /*  /dev/md0 -> 0
 602                  *  /dev/md_d0 -> d0
 603                  *  /dev/md/1 -> 1
 604                  *  /dev/md/d1 -> d1
 605                  *  /dev/md/home -> home
 606                  *  /dev/mdhome -> home
 607                  */
 608                 /* FIXME compare this with rules in create_mddev */
 609                 name = strrchr(mddev, '/');
 610                 if (name) {
 611                         name++;
 612                         if (strncmp(name, "md_d", 4)==0 &&
 613                             strlen(name) > 4 &&
 614                             isdigit(name[4]) &&
 615                             (name-mddev) == 5 /* /dev/ */)
 616                                 name += 3;
 617                         else if (strncmp(name, "md", 2)==0 &&
 618                                  strlen(name) > 2 &&
 619                                  isdigit(name[2]) &&
 620                                  (name-mddev) == 5 /* /dev/ */)
 621                                 name += 2;
 622                 }
 623         }
 624         if (!st->ss->init_super(st, &info.array, size, name, homehost, uuid))
 625                 goto abort;
 626
 627         total_slots = info.array.nr_disks;
 628         sysfs_init(&info, mdfd, 0);
 629         st->ss->getinfo_super(st, &info, NULL);
 630
 631         if (did_default && verbose >= 0) {
 632                 if (is_subarray(info.text_version)) {
 633                         int dnum = devname2devnum(info.text_version+1);
 634                         char *path;
 635                         int mdp = get_mdp_major();
 636                         struct mdinfo *mdi;
 637                         if (dnum > 0)
 638                                 path = map_dev(MD_MAJOR, dnum, 1);
 639                         else
 640                                 path = map_dev(mdp, (-1-dnum)<< 6, 1);
 641
 642                         mdi = sysfs_read(-1, dnum, GET_VERSION);
 643
 644                         fprintf(stderr, Name ": Creating array inside "
 645                                 "%s container %s\n",
 646                                 mdi?mdi->text_version:"managed", path);
 647                         sysfs_free(mdi);
 648                 } else
 649                         fprintf(stderr, Name ": Defaulting to version"
 650                                 " %s metadata\n", info.text_version);
 651         }
 652
 653         map_update(&map, fd2devnum(mdfd), info.text_version,
 654                    info.uuid, chosen_name);
 655         map_unlock(&map);
 656
 657         if (bitmap_file && vers < 9003) {
 658                 major_num = BITMAP_MAJOR_HOSTENDIAN;
 659 #ifdef __BIG_ENDIAN
 660                 fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n"
 661                         "  between different architectured.  Consider upgrading the Linux kernel.\n");
 662 #endif
 663         }
 664
 665         if (bitmap_file && strcmp(bitmap_file, "internal")==0) {
 666                 if ((vers%100) < 2) {
 667                         fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n");
 668                         goto abort;
 669                 }
 670                 if (!st->ss->add_internal_bitmap) {
 671                         fprintf(stderr, Name ": internal bitmaps not supported with %s metadata\n",
 672                                 st->ss->name);
 673                         goto abort;
 674                 }
 675                 if (!st->ss->add_internal_bitmap(st, &bitmap_chunk,
 676                                                  delay, write_behind,
 677                                                  bitmapsize, 1, major_num)) {
 678                         fprintf(stderr, Name ": Given bitmap chunk size not supported.\n");
 679                         goto abort;
 680                 }
 681                 bitmap_file = NULL;
 682         }
 683
 684
 685         sysfs_init(&info, mdfd, 0);
 686
 687         if (st->ss->external && st->container_dev != NoMdDev) {
 688                 /* member */
 689
 690                 /* When creating a member, we need to be careful
 691                  * to negotiate with mdmon properly.
 692                  * If it is already running, we cannot write to
 693                  * the devices and must ask it to do that part.
 694                  * If it isn't running, we write to the devices,
 695                  * and then start it.
 696                  * We hold an exclusive open on the container
 697                  * device to make sure mdmon doesn't exit after
 698                  * we checked that it is running.
 699                  *
 700                  * For now, fail if it is already running.
 701                  */
 702                 container_fd = open_dev_excl(st->container_dev);
 703                 if (container_fd < 0) {
 704                         fprintf(stderr, Name ": Cannot get exclusive "
 705                                 "open on container - weird.\n");
 706                         goto abort;
 707                 }
 708                 if (mdmon_running(st->container_dev)) {
 709                         if (verbose)
 710                                 fprintf(stderr, Name ": reusing mdmon "
 711                                         "for %s.\n",
 712                                         devnum2devname(st->container_dev));
 713                         st->update_tail = &st->updates;
 714                 } else
 715                         need_mdmon = 1;
 716         }
 717         rv = set_array_info(mdfd, st, &info);
 718         if (rv) {
 719                 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
 720                         mddev, strerror(errno));
 721                 goto abort;
 722         }
 723
 724         if (bitmap_file) {
 725                 int uuid[4];
 726
 727                 st->ss->uuid_from_super(st, uuid);
 728                 if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk,
 729                                  delay, write_behind,
 730                                  bitmapsize,
 731                                  major_num)) {
 732                         goto abort;
 733                 }
 734                 bitmap_fd = open(bitmap_file, O_RDWR);
 735                 if (bitmap_fd < 0) {
 736                         fprintf(stderr, Name ": weird: %s cannot be openned\n",
 737                                 bitmap_file);
 738                         goto abort;
 739                 }
 740                 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
 741                         fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
 742                                 mddev, strerror(errno));
 743                         goto abort;
 744                 }
 745         }
 746
 747         infos = malloc(sizeof(*infos) * total_slots);
 748
 749         for (pass=1; pass <=2 ; pass++) {
 750                 struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */
 751
 752                 for (dnum=0, dv = devlist ; dv ;
 753                      dv=(dv->next)?(dv->next):moved_disk, dnum++) {
 754                         int fd;
 755                         struct stat stb;
 756                         struct mdinfo *inf = &infos[dnum];
 757
 758                         if (dnum >= total_slots)
 759                                 abort();
 760                         if (dnum == insert_point) {
 761                                 moved_disk = dv;
 762                                 continue;
 763                         }
 764                         if (strcasecmp(dv->devname, "missing")==0)
 765                                 continue;
 766                         if (have_container)
 767                                 moved_disk = NULL;
 768                         if (have_container && dnum < info.array.raid_disks - 1)
 769                                 /* repeatedly use the container */
 770                                 moved_disk = dv;
 771
 772                         switch(pass) {
 773                         case 1:
 774                                 *inf = info;
 775
 776                                 inf->disk.number = dnum;
 777                                 inf->disk.raid_disk = dnum;
 778                                 if (inf->disk.raid_disk < raiddisks)
 779                                         inf->disk.state = (1<<MD_DISK_ACTIVE) |
 780                                                 (1<<MD_DISK_SYNC);
 781                                 else
 782                                         inf->disk.state = 0;
 783
 784                                 if (dv->writemostly == 1)
 785                                         inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
 786
 787                                 if (have_container)
 788                                         fd = -1;
 789                                 else {
 790                                         if (st->ss->external &&
 791                                             st->container_dev != NoMdDev)
 792                                                 fd = open(dv->devname, O_RDWR);
 793                                         else
 794                                                 fd = open(dv->devname, O_RDWR|O_EXCL);
 795
 796                                         if (fd < 0) {
 797                                                 fprintf(stderr, Name ": failed to open %s "
 798                                                         "after earlier success - aborting\n",
 799                                                         dv->devname);
 800                                                 goto abort;
 801                                         }
 802                                         fstat(fd, &stb);
 803                                         inf->disk.major = major(stb.st_rdev);
 804                                         inf->disk.minor = minor(stb.st_rdev);
 805                                 }
 806                                 if (fd >= 0)
 807                                         remove_partitions(fd);
 808                                 if (st->ss->add_to_super(st, &inf->disk,
 809                                                          fd, dv->devname)) {
 810                                         ioctl(mdfd, STOP_ARRAY, NULL);
 811                                         goto abort;
 812                                 }
 813                                 st->ss->getinfo_super(st, inf, NULL);
 814                                 safe_mode_delay = inf->safe_mode_delay;
 815
 816                                 if (have_container && verbose > 0)
 817                                         fprintf(stderr, Name ": Using %s for device %d\n",
 818                                                 map_dev(inf->disk.major,
 819                                                         inf->disk.minor,
 820                                                         0), dnum);
 821
 822                                 if (!have_container) {
 823                                         /* getinfo_super might have lost these ... */
 824                                         inf->disk.major = major(stb.st_rdev);
 825                                         inf->disk.minor = minor(stb.st_rdev);
 826                                 }
 827                                 break;
 828                         case 2:
 829                                 inf->errors = 0;
 830                                 rv = 0;
 831
 832                                 rv = add_disk(mdfd, st, &info, inf);
 833
 834                                 if (rv) {
 835                                         fprintf(stderr,
 836                                                 Name ": ADD_NEW_DISK for %s "
 837                                                 "failed: %s\n",
 838                                                 dv->devname, strerror(errno));
 839                                         goto abort;
 840                                 }
 841                                 break;
 842                         }
 843                         if (!have_container &&
 844                             dv == moved_disk && dnum != insert_point) break;
 845                 }
 846                 if (pass == 1) {
 847                         struct mdinfo info_new;
 848                         struct map_ent *me = NULL;
 849
 850                         /* check to see if the uuid has changed due to these
 851                          * metadata changes, and if so update the member array
 852                          * and container uuid.  Note ->write_init_super clears
 853                          * the subarray cursor such that ->getinfo_super once
 854                          * again returns container info.
 855                          */
 856                         map_lock(&map);
 857                         st->ss->getinfo_super(st, &info_new, NULL);
 858                         if (st->ss->external && level != LEVEL_CONTAINER &&
 859                             !same_uuid(info_new.uuid, info.uuid, 0)) {
 860                                 map_update(&map, fd2devnum(mdfd),
 861                                            info_new.text_version,
 862                                            info_new.uuid, chosen_name);
 863                                 me = map_by_devnum(&map, st->container_dev);
 864                         }
 865
 866                         st->ss->write_init_super(st);
 867
 868                         /* update parent container uuid */
 869                         if (me) {
 870                                 char *path = strdup(me->path);
 871
 872                                 st->ss->getinfo_super(st, &info_new, NULL);
 873                                 map_update(&map, st->container_dev,
 874                                            info_new.text_version,
 875                                            info_new.uuid, path);
 876                                 free(path);
 877                         }
 878                         map_unlock(&map);
 879
 880                         flush_metadata_updates(st);
 881                         st->ss->free_super(st);
 882                 }
 883         }
 884         free(infos);
 885
 886         if (level == LEVEL_CONTAINER) {
 887                 /* No need to start.  But we should signal udev to
 888                  * create links */
 889                 sysfs_uevent(&info, "change");
 890                 if (verbose >= 0)
 891                         fprintf(stderr, Name ": container %s prepared.\n", mddev);
 892                 wait_for(chosen_name, mdfd);
 893         } else if (runstop == 1 || subdevs >= raiddisks) {
 894                 if (st->ss->external) {
 895                         int err;
 896                         switch(level) {
 897                         case LEVEL_LINEAR:
 898                         case LEVEL_MULTIPATH:
 899                         case 0:
 900                                 err = sysfs_set_str(&info, NULL, "array_state",
 901                                                     "active");
 902                                 need_mdmon = 0;
 903                                 break;
 904                         default:
 905                                 err = sysfs_set_str(&info, NULL, "array_state",
 906                                                     "readonly");
 907                                 break;
 908                         }
 909                         sysfs_set_safemode(&info, safe_mode_delay);
 910                         if (err) {
 911                                 fprintf(stderr, Name ": failed to"
 912                                         " activate array.\n");
 913                                 ioctl(mdfd, STOP_ARRAY, NULL);
 914                                 goto abort;
 915                         }
 916                 } else {
 917                         /* param is not actually used */
 918                         mdu_param_t param;
 919                         if (ioctl(mdfd, RUN_ARRAY, &param)) {
 920                                 fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
 921                                         strerror(errno));
 922                                 ioctl(mdfd, STOP_ARRAY, NULL);
 923                                 goto abort;
 924                         }
 925                 }
 926                 if (verbose >= 0)
 927                         fprintf(stderr, Name ": array %s started.\n", mddev);
 928                 if (st->ss->external && st->container_dev != NoMdDev) {
 929                         if (need_mdmon)
 930                                 start_mdmon(st->container_dev);
 931
 932                         ping_monitor(devnum2devname(st->container_dev));
 933                         close(container_fd);
 934                 }
 935                 wait_for(chosen_name, mdfd);
 936         } else {
 937                 fprintf(stderr, Name ": not starting array - not enough devices.\n");
 938         }
 939         close(mdfd);
 940         return 0;
 941
 942  abort:
 943         map_lock(&map);
 944         map_remove(&map, fd2devnum(mdfd));
 945         map_unlock(&map);
 946
 947         if (mdfd >= 0)
 948                 close(mdfd);
 949         return 1;
 950 }